EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Reduce CI duration of test_lora_update. (#9024)

Browse Source
This commit is contained in:
Lifu Huang
2025-08-10 15:34:04 -07:00
committed by GitHub
parent 2c7f01bc89
commit e322a94d1f
2 changed files with 49 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
test/srt/lora/test_lora_update.py
View File

@@ -229,6 +229,53 @@ BASIC_TESTS = [
]
),
),
Operation(
type=OperationType.UNLOAD,
data="Nutanix/Meta-Llama-3.1-8B-Instruct_lora_4_alpha_16",
),
Operation(
type=OperationType.UNLOAD,
data="pbevan11/llama-3.1-8b-ocr-correction",
),
Operation(
type=OperationType.FORWARD,
data=create_batch_data(
"Nutanix/Meta-Llama-3.1-8B-Instruct_lora_4_alpha_16"
),
expected_error="not loaded",
),
Operation(
type=OperationType.FORWARD,
data=create_batch_data("pbevan11/llama-3.1-8b-ocr-correction"),
expected_error="not loaded",
),
Operation(
type=OperationType.FORWARD,
data=create_batch_data(None),
),
Operation(
type=OperationType.LOAD,
data="philschmid/code-llama-3-1-8b-text-to-sql-lora",
),
Operation(
type=OperationType.LOAD,
data="Nutanix/Meta-Llama-3.1-8B-Instruct_lora_4_alpha_16",
),
Operation(
type=OperationType.LOAD,
data="pbevan11/llama-3.1-8b-ocr-correction",
),
Operation(
type=OperationType.FORWARD,
data=create_batch_data(
[
"philschmid/code-llama-3-1-8b-text-to-sql-lora",
"Nutanix/Meta-Llama-3.1-8B-Instruct_lora_4_alpha_16",
"pbevan11/llama-3.1-8b-ocr-correction",
None,
]
),
),
],
),
]
@@ -1194,7 +1241,7 @@ class TestLoRADynamicUpdate(CustomTestCase):
"""
Test dynamic LoRA updates in engine mode.
"""
test_cases = ALL_TESTS
test_cases = BASIC_TESTS if is_in_ci() else ALL_TESTS
self._run_dynamic_adapter_updates(
mode=LoRAUpdateTestSessionMode.ENGINE,
test_cases=test_cases,
@@ -1204,9 +1251,7 @@ class TestLoRADynamicUpdate(CustomTestCase):
"""
Test dynamic LoRA updates in server mode.
"""
# In CI, we only run the first test case to save time, as the engine test should be mostly sufficient for ensuring correctness.
test_cases = BASIC_TESTS if is_in_ci() else ALL_TESTS
self._run_dynamic_adapter_updates(
mode=LoRAUpdateTestSessionMode.SERVER, test_cases=test_cases
)

2
test/srt/run_suite.py
View File

@@ -21,7 +21,7 @@ suites = {
TestFile("lora/test_lora_backend.py", 99),
TestFile("lora/test_multi_lora_backend.py", 60),
TestFile("lora/test_lora_cuda_graph.py", 250),
TestFile("lora/test_lora_update.py", 800),
TestFile("lora/test_lora_update.py", 400),
TestFile("lora/test_lora_qwen3.py", 97),
TestFile("models/test_embedding_models.py", 73),
# TestFile("models/test_clip_models.py", 52),